The Origins and Future of Distributed Computing and Clouds - The Practice of Cloud System Administration - page 455

For high availability, sites still needed three machines and the resulting 66 percent idle
capacity.
Costs
The cost of providing a highly available and popular web service during this era was very
high. It was run on expensive high-end hardware, with expensive reliability features such
as RAID and multiple CPUs, in an N + 1 or N + 2 configuration. That architecture meant
that 50 to 66 percent of this expensive hardware was always idle. To reduce downtime to
next to nothing, sites might also use expensive load balancers.
The OS costs and support costs for this high-end hardware were also very high, as were
the costs for the application software, such as the web server software and database soft-
ware.
Scaling a site built in this way meant that the cost per unit of performance increased
as the site got bigger. Rather than more customers resulting in economies of scale, more
customers meant a higher cost per unit. The cost model was the reverse of what a business
would normally expect.
B.3 The Dot-Bomb Era (2000-2003)
The dot-com bubble collapse started in 2000. By 2001, the bubble was deflating as fast as
it could. Most of the dot-coms burned through their venture capital and their stock ceased
trading, often never having reached profitability.
After the collapse came a period of calm. Things slowed down a bit. It was possible to
pause and consider what had been learned from the past. Without all the hype, marketing,
andeasymoney,thebettertechnologiessurvived.Withoutpressurefrominvestorswanting
their cash spent quickly so that they would get a quick return on their investment, people
could take enough time to think and invent new solutions. Silicon Valley is, for the most
part, a meritocracy. Working technology rules; hype and self-promotion are filtered out.
Availability Requirements
During the dot-bomb era, there were no significant changes in availability requirements.
TheInternet-basedcompaniesthathadsurvivedthecrashdevelopedabetterunderstanding
of their availability requirements and figured out how to meet them without breaking the
bank.
Technology
Three trends enabled the next phase: surplus capacity left over from the previous boom
years, the commoditization of hardware, and the maturation of open source software.